Glossary
On-device recognition
Speech converted to text on your Mac, with no server involved.
What it means
On-device recognition runs the speech model locally — the audio is transcribed by your own machine and never crosses the network. The contrast is cloud dictation, where audio is streamed to a service and returned as text.
Why it matters
Voice is biometric data said out loud, often in rooms other people can hear. On-device is the only answer to “where does my dictation go?” that survives the question being asked twice.
How notchos does it
Recognition runs through Apple’s Speech framework with audio buffers discarded the moment a session ends — audio never touches disk. History stays local, with a retention window you choose.
Fair questions
- Is on-device recognition as accurate as cloud?
- For dictation of prose it is excellent on modern macOS; the deterministic cleanup pass then fixes what recognition leaves rough.
- Does it work offline?
- Yes — with the on-device recognizer there is no network dependency at all.
Related terms
Twenty tools. One hover. $9.99, once.